A Hidden Markov Model Development for Star Fruit Sugariness Identification

Abstract

This research is addressed the implementation of Hidden Markov Model (HMM) dedicated to identify the star fruit sugariness. This pixel based identification transforms the RGB (Red-Green-Blue) image into the total dissolved solids (TDS) for each pixel by using a linear regression equation, and categorized them into one of the six available Class. Then the sequence of pixels with one out of six available symbols becomes the input sequence. Based on this sequence, we identify the star fruit sugariness into one of the three categories by using HMM. Number of hidden state is varying frem 3 until 8. This research shown that is a positive correlation between TDS with the Red component, and negative correlation with Green component. There is non significant correlation between TDS with the blue component The best regression model that relating TDS with the RGB component is TDS=f(R,G) with 0.85 as determination coefficients (R2). The accuracy of the HMM is varies from 69% to 75 %, with 73.5% as average. It seen that the classification is tend to the low class. It means, the misclassifications occur from class 2 classified into class 1, and class 3 classified into class 2. All objects in class 1 are classified into appropriate class.